Anthropic has added an extension mechanism called mods to Claude Code. By writing a small amount of TypeScript, developers can rewrite its prompts, add new interface elements, or even replace a built-in feature.
Mods are distributed with plugins and can be used in both command line and desktop versions. The official has changed some of the built-in functions into the form of mods, and will continue to migrate them in the future. In other words, part of the "ontology" of Claude Code will become modules that can be replaced in the future.
There is also an interesting usage: you can directly ask Claude to write mods for you. Describe what kind of status bar and what kind of shortcut commands you want, and it will generate and install the code itself.
Please read the risk warning carefully. Mods have the same permissions as Claude Code, without sandbox isolation. What a mod can do is equivalent to everything Claude Code can do on your computer - read files, run commands, and access the network. The official official words mean: Only install sources you trust.
Version 2.1.323, released on the same day, also has an experimental Code Mode that lets Claude use JavaScript to call tools, and organization administrators can turn it off. Plug-in management has also been tightened: plug-ins that do not match the declared version range will not be installed or updated, and plug-in validation has also begun to check the type of TypeScript hooks.
From a product perspective, programming intelligence is moving from “a tool” to “a platform”. As customizability increases, supply chain security issues will also arise. Before the team uses it, it is best to make a list of plug-ins that are allowed to be installed internally.